Analyzing And Modeling Human Spatio-Temporal Behaviors With Mobile Data

With the evolution of wireless technologies,a bunch of mobile applications accumulate massive amount of data which conveys plentiful information about social behavior,use habits or personal preferences etc.Some of these datasets record directly the coordinates of whereabouts and engaging behavior,while others encode indirectly physical surroundings and social connections.By employing a passive collection technique,we in this paper obtain multi-source mobile network data under varying spatial granularities,and then perform analyses and modelling on the patterns and rules in human spatio-temporal behavior,which considers behavioral properties from both individual and group aspects simultaneously.The systematical approaches for quantification,analyzing and modelling of human behavior have potential values in multiple R&D areas,e.g.the epidemic prediction,urban management,and mobile network optimization.Meanwhile,our researches also make contributions for the theoretical development of behavioral and network sciences.Specifically,we summarize our achievements as following:We propose a novel framework to evaluate the quality of spatio-temporal datasets,as well as a quality-enhancing algorithm for human mobility.This evaluation method focuses on the common form and properties in spatio-temporal data,by considering objective metrics for data quality from single data point,individual and group perspectives,respectively.For a single data point,we quantify the data quality with the static spatio-temporal resolution and dynamic transitions between successive records.For an individual's trajectory,we compress the quality of multiple data points into a single metric by calculating the heterogeneity of a sequence of spatiotemporal observations.This method shows remarkable performance efficiency when comparing with traditional entropy metric.For trajectories of a group of people,we involve space splitting to combine the correlation between different blocks and feature distribution within the same block.We finally apply our evaluation on the real datasets,and propose a quality-enhancing method for human mobility data,which shows a great performance improvement when comparing with existing spatial or temporal interpolations.We have studied the unified spatio-temporal mesostructures in human mobility.To discover the mesostructures from cellular data,we first introduce a topology-attributes coupling similarity algorithm to derive the elementary(nodes and edges)similarities for two attributed graphs.With the construction of individual profiles from mesostructure analyses,we provided a novel mobility model from a process-driven perspective,which reduced the dependence of many existing models on the consistency between local and global mobility statistics.We gained some insights on the dominating mesostructures in human mobility by leveraging mobile data in a large city.The statistical distribution of mesostructures is found to be determined by the intrinsic heterogeneity of spatio-temporal properties in human behavior.Our model evaluation showed that a process with basic rules could demonstrate the key statistical properties in mobility mesostructures.We believe that these approaches and observations would be a good reference for management of human mobility in mobile networks and transportation systems.Beyond the spatio-temporal dependence in individual's mobility as revealed by mesostructures,we also investigate the dependence at population level.Despite recent progress in revealing temporal dynamics and spatial inhomogeneity of group mobility,limited knowledge about spatio-temporal dependence is gained.One of challenges comes from the absence of sustained observations at varying spatial scales.We characterize the group dynamics with correlation functions and measures the spatial and temporal properties statistically.Different from previous observations with single data source,we compare the group mobility dynamics with three varying granularities,i.e.,campus,city and country.We eventually model the spatio-temporal dependence,whose evaluation results suggest connections between spatio-temporal dependence of group mobility and the organization of human lives.Region differences and spatial scales are observed to impact spatio-temporal dependence to a great extent.Additionally,interactive knowledge between space and time enhances population prediction with a decrease in rootmean-square error of 2.8%?25.2%.We believe that these achievements will benefit multiple research and development areas such as network deploying and simulation researches.Although passive measurements of mobile traffic have been conducted in previous literature,they mostly address protocol and traffic properties,rather than responsive user behaviour consequences.In this paper,we perform a characterization of mobile traffic and engaging behaviours from end-user's view.The proposed concurrence index equipped by the model is more powerful to capture delicate difference of user-perceived application performance than previous volume-based metrics.Then we profile the behavioural dynamics of user participation in mobile usage and its interaction with user-perceived application performance.And finally,we perform a unique modelling of individual engaging trajectories and a model-based clustering to explore user behavioural patterns.We find that user engaging behaviour is primarily governed by a small portion of latent states,and the behavioural patterns regarding principle engaging states illustrate distinctive properties in discovered user clusters.